PyTorch implementation of the spectral parametrization of the weights of a 2D convolutional layer, as introduced in the paper "Efficient Nonlinear Transforms for Lossy Image Compression" by Johannes Ballé, PCS 2018.
